Anthropogenic pressures coincide with Neotropical biodiversity hotspots in a flagship butterfly group

The biodiversity crisis has highlighted the need to assess and map biodiversity in order to prioritize conservation efforts. Clearwing butterflies (tribe Ithomiini) have been proposed as biological indicators for habitat quality in Neotropical forests, which contain the world's richest biological communities. Here, we provide maps of different facets of Ithomiini diversity across the Neotropics to identify areas of evolutionary and ecological importance for conservation and evaluate their overlap with current anthropogenic threats.


| INTRODUC TI ON
The global biodiversity crisis is a critical environmental issue (IPBES, 2019) with unprecedented rates of species loss across multiple taxonomic groups, now referred to as the sixth mass extinction (Barnosky et al., 2011;Ceballos et al., 2015). Such species loss significantly alters biodiversity patterns and affects ecosystem functions worldwide. In particular, there is growing evidence for recent massive declines in insects (Cardoso et al., 2020;Eggleton, 2020;Montgomery et al., 2020), which represent the bulk of current biodiversity (Mora et al., 2011;Stork, 2018). This loss is concomitant with the global increase in human pressures on ecosystems, with currently 75% of the planet's non-frozen land surface impacted (Venter et al., 2016b). These alarming trends are compelling scientists to better assess and map biodiversity in order to prioritize conservation efforts given limited time and resources (Brooks et al., 2006).
One early approach towards identifying global priority areas for conservation was the delimitation of biodiversity hotspots (Myers et al., 2000): areas with high levels of vascular plant species richness and endemism, and significant loss of primary natural habitats.
The Neotropics encompass seven of these biodiversity hotspots, including the richest of them: the Tropical Andes (Myers et al., 2000). However, it is not known how well such hotspots, identified on the basis of vascular plant diversity and confirmed for vertebrates, provide adequate surrogates for the diversity of other taxa, especially insects (Stork & Habel, 2014). Indeed, georeferenced primary biodiversity data for insects, particularly in the Neotropics, are very scarce due to the challenges of collecting, digitalizing and verifying taxonomic identifications for records covering often inaccessible, remote regions (Short et al., 2018;Stork, 2018). Moreover, whether species richness and endemism adequately reflect other facets of biodiversity such as phylogenetic and functional diversity may depend on the group considered (Albouy et al., 2017;Allouche et al., 2006;Devictor et al., 2010;Mazel et al., 2014;Prendergast et al., 1993;Williams et al., 1996;Zupan et al., 2014). There is therefore an urgent need to explore to what extent existing hotspots identified for well-studied taxa coincide with those of other less well-known groups, and how well those hotspots represent facets of biodiversity beyond species richness and endemism. Here, we tackle this issue by investigating the spatial distribution of different metrics of biodiversity in an integrative assessment that covers its multifaceted nature (Pollock et al., 2017(Pollock et al., , 2020. We focus our assessment on a diverse insect group, the butterfly tribe Ithomiini Godman & Salvin, 1879 (Nymphalidae: Danainae), in the world's biologically richest region, the Neotropics.
The tribe Ithomiini comprises 396 described species distributed among 42 genera and 10 subtribes (Chazot et al., 2019). These butterflies form diverse communities in humid forests from sea level to 3000 m, throughout the Neotropics. Their habitats are threatened by high rates of deforestation associated with cattle ranches, soybean and oil palm plantations, as well as industrial logging, mining and road building (Armenteras et al., 2017;Fearnside, 2017;Rajão et al., 2020;Sonter et al., 2017). In this context, ithomiine butterflies have been proposed as indicator species for habitat quality and local butterfly diversity (Beccaloni & Gaston, 1995;Brown Jr, 1997;Uehara-Prado & Freitas, 2009 but see Brown & Freitas, 2000). Ithomiini also represent the most diverse radiation of aposematic and Müllerian mimetic butterflies, whereby co-occurring unpalatable species display similar wing colour patterns that advertise their distastefulness to predators. Müllerian co-mimetic species interact mutualistically, because they share the cost incurred during the learning process of predators (Joron & Mallet, 1998;Müller, 1879;Sherratt, 2008). All Ithomiini species engage in Müllerian mimicry and drive mimicry in other distantly related groups of Lepidoptera (Beccaloni, 1997;Brown Jr. & Benson, 1974). Remarkably, many ithomiines have partly transparent wings (McClure et al., 2019;Papageorgis, 1975;Figure 1a), which has inspired their common name of 'clearwing' butterflies. Overall, ithomiine butterflies combine their potential role as biological indicators with positive public image (e.g. Barua et al., 2012;Sumner et al., 2018), making them candidate flagship species for conservation in the Neotropics. Previous studies have already investigated the historical biogeography of the tribe. Ithomiini likely originated in the eastern Andean foothills and a major clade, composed of the five most species-rich subtribes and comprising 80% of species, diversified in Central Andes 20-10 My ago (Chazot et al., 2019). Those areas, which harbour heterogeneous landscapes favouring speciation, also coincide with known hotspots of species richness for three diverse ithomiine genera (Ithomia, Napeogenes, and Oleria;. However, patterns of species diversity remain to be documented at the level of the entire tribe, across the Neotropics. Similarly, patterns of phylogenetic diversity, geographic rarity and mimicry richness remain largely uncharacterized at such scales. Phylogenetic diversity has become a fundamental component of biodiversity assessments that addresses the evolutionary distinctiveness of species assemblages (Faith, 1992). It is recognized by the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) as a key indicator for the maintenance of options in nature's contribution for people (IPBES, 2019) and is included in conservation tools such as the EDGE index for Evolutionarily Distinct and Globally Endangered species (Isaac et al., 2007). High phylogenetic diversity reflects the presence of species with distinctive evolutionary pathways, which provide a proxy for evolutionary novelties of high value for conservation (Faith, 2018).
Geographic rarity is another component of biodiversity that relates to the spatial dimension of rarity, with species with restricted distributions being considered rare compared to species with wider ranges (Rabinowitz, 1981). Species with small geographic ranges can support original functions in ecosystems (Mouillot et al., 2013), while they often face higher risks of extinction Cardillo et al., 2008;Purvis et al., 2000). Species range size is therefore commonly incorporated into diversity indices (Gumbs et al., 2020;Jetz et al., 2014;Maritz et al., 2016). Mapping species geographic rarity provides an additional tool for conservation prioritization (Cadotte & Davies, 2010), as another complementary facet of diversity patterns linked to species vulnerability and areas of endemism.
Biotic interactions, although rarely integrated in biodiversity assessments, represent the architecture of life that reflects the interdependence of all units of biodiversity (Bascompte, 2009). For example, mutualistic interactions can shape species distributions and community composition (Duffy & Johnson, 2017;Sherratt, 2006), affecting ecosystem stability (Hale et al., 2020;Pascual-García & Bastolla, 2017) and supporting ecosystem services such as pollination function or seed dispersal (Millennium Ecosystem Assessment,   (Aubier et al., 2017;Chazot et al., 2014;Kapan, 2001;Langham, 2004;Mallet & Barton, 1989;Sherratt, 2006). Additionally, mimetic species form adaptively assembled mutualistic communities that are predicted to suffer more from community disassembly due to the loss of those mutualistic interactions (Toby Kiers et al., 2010) and to be more sensitive to co-extinction cascades (Dunn et al., 2009). Therefore, Müllerian mimicry systems provide opportunities to map patterns of mimicry richness and geographic rarity, which reflect the distribution of mutualistic interactions in space, a component of functional diversity that is particularly relevant for conservation.
In this study, (1) we provide modelled distribution maps of taxonomic, phylogenetic and mimetic diversity as well as geographic rarity, for the entire tribe Ithomiini across the Neotropics, in order to identify biodiversity hotspots as areas of both evolutionary and ecological importance for conservation; (2) we evaluate the spatial relationships among those facets of Ithomiini diversity; and (3) we assess current anthropogenic threats to Ithomiini biodiversity hotspots, highlighting risk areas with high anthropogenic pressures, and potential refuges with currently low levels of human influence.

| Data sources
We compiled from multiple sources an initial data set of 28,986 georeferenced occurrences for 388 ithomiine butterfly species in their natural habitats, out of the 396 known species, spanning 25 countries across the Neotropics (see maps of occurrences, sampling effort, sampling completeness and bioregions in Figure S1.1, S1.2, S1.3 & S1.4). This data set provided 19,271 species-grid-cell records for distribution modelling at a 0.25° × 0.25° spatial resolution after removing duplicate records from single grid cells, which are available from Zenodo at https://doi.org/10.5281/zenodo.4696055. The data come from fieldwork by the authors over the past five decades, and records from over 60 museums and private collections detailed in the online archive metadata. Each record is associated with its location, its taxonomic identity and its mimicry ring membership (i.e. a wing colour pattern shared by individuals reflecting mutualistic interactions). The current classification of wing patterns presents 44 mimicry rings ( Figure S2.5) updated from previous works (Beccaloni, 1997;Chazot et al., 2014Chazot et al., , 2019Elias et al., 2008;Jiggins et al., 2006;Willmott & Mallet, 2004).
Sets of co-mimetic species (i.e. sharing a wing pattern) form mimicry rings (Figure 1b) Figure S3.6), which represents 339 out of the 388 species with georeferenced records.

| Species distribution modelling (SDM)
In order to map the current distributions of ithomiines, we developed species distribution models (SDMs) relating occurrence data with a set of environmental variables. We describe our SDM methods following the ODMAP (Overview, Data, Model, Assessment, Prediction) protocol for species distribution models (Zurell et al., 2020). Here, we provide the overview of the distribution models while the remaining ODMAP sections, providing details in modelling steps, justifications for modelling choices and a more in-depth discussion about potential caveats and limits, are detailed in Appendix

5.
We aimed to model the current distribution of species and mimicry rings, as well as to infer the current patterns of geographic rarity and taxonomic, phylogenetic and mimicry ring diversity (as described further below) for the whole Ithomiini tribe. We proceeded as follows: (1) we retrieved environmental predictors of Ithomiini distribution, (2) we obtained multiple environmental suitability maps for each OMU employing a set of SDM algorithms, (3) we derived for each OMU a median ensemble model depicting its modelled distribution, (4) we stacked these modelled distribution maps in order to obtain in each pixel the predicted occurrence of OMUs, while observed binary maps of OMU with less than six occurrences were included directly at this step, then we (5) derived species and mimicry ring distribution maps and compute various taxonomic, phylogenetic and mimicry diversity and geographic rarity indices ( Figure 2 i.e. 86.3% of species with known occurrences). We included the remaining 220 OMUs (28.1%) in stacks as binary rasters of presencesabsences. We fitted SDMs in biomod2 v.3.4.6  using three machine learning algorithms to cope with small sample sizes: random forest, generalized boosted models and artificial neural networks. We drew pseudo-absences from those occurrences in a target group strategy (Mateo et al., 2010), a procedure to increase the likelihood that sampled pseudo-absences were effectively located in sites where OMUs were absent. We evaluated model performance with maximized Jaccard indices. For 361 OMUs with small sample sizes (N < 30; 46.1%), models were evaluated upon the calibration data set. For 202 OMUs with large sample sizes (N ≥ 30; 25.8%), we ran an additional 3-fold spatial block cross-validation step (Roberts et al., 2017;Valavi et al., 2019) to improve model evaluation. We discarded models with a poor performance (Jaccard index < 0.95 without cross-validation; Jaccard index < 0.6 with cross-validation) and produced an ensemble model based on the median of predictions. We clipped final outputs with OMU-specific buffered alpha-hulls and, where relevant, we constrained outputs to the east or west of the Andean continental divide, to limit the extent of possible distributions to reasonable areas.
We derived species and mimicry ring distribution maps from the modelled distribution maps of OMUs as the likelihood of finding at least one of the OMUs belonging to the species/mimicry ring in the community (i.e. in the pixel). In the final post-processing step, we computed six diversity and geographic rarity indices based directly on the stack of species and mimicry ring maps. Additionally, we present in Appendix 9 four additional indices evaluating similar facets of diversity with alternative methods to explore the robustness of our analyses to index selection.

| Diversity and geographic rarity indices
We computed species and mimicry richness as the expected number of species and mimicry rings found in our communities (i.e. in each pixel), by summing the continuous outputs from models as recommended by Calabrese et al. (2014). To estimate phylogenetic diversity, we computed Faith's phylogenetic diversity index (Faith, 1992) based on the phylogeny of the Ithomiini tribe (Chazot et al., 2019) encompassing 339 species and 719 OMUs. This index estimates the total length of branches connecting all the species within a community, capturing the quantity of evolutionary history they represent.
We assigned geographic rarity weights for each species and mimicry ring based on their relative geographic ranges following the threshold-dependent exponentially decaying weighting scheme of Leroy et al. (2013). This method assigns weights that exponentially increase below the chosen rarity threshold and rapidly decay to zero above the threshold, thereby limiting the impact of common species on community indices. We chose the rarity threshold at which the average proportion of rare species in communities was 25%, as detailed in Leroy et al. (2012). Next, we used these rarity weights to calculate an index of rarity for each community, which was the average rarity weight for all species or mimicry rings. These indices F I G U R E 2 Species Distribution Model (SDM) workflow depicting the different analytical steps performed. Distribution models are computed for each OMU. Depending on sample size, modelling steps and settings differed. Clipping step to constrain SDM projections to plausible distribution ranges is not shown on the chart. Algo = algorithms used in the study, namely random forest (RF), gradient boosting models (GBM) and artificial neural networks (ANN); PAs = sets of pseudo-absences; CV = cross-validation folds; mim. rings = mimicry rings can be seen as proxies for relative levels of endemism since they quantify the relative importance of species or mimicry ring with small ranges in communities.
To quantify the importance of mutualistic interactions, we estimated the mean size (i.e. number of species) for mimicry rings within each community. Communities with high mean mimicry ring size correspond to greater frequencies of mutualistic interactions, while communities with low mean mimicry ring size host in average species engaged in fewer mutualistic interactions. Assuming that the richest mimicry rings also tend to be the most abundant, species belonging to smaller mimicry rings, thus harbouring locally rare patterns, are likely more vulnerable to predation by naïve predators, and thereby to local extinctions (Müller, 1879). As such, a low mean mimicry ring size may relate to higher vulnerability on average in the mimicry community.
Additionally, we computed indices of effective richness based on Shannon's diversity indices and an index of evolutionary distinctiveness based on Fair-Proportions (Redding, 2003), and we mapped the size of the main mimicry ring in each community (see Figure S9.22).
A flow chart and additional details on index computation based on our modelled distribution maps are provided in Appendix 8. The robustness of indices was tested with several sensitivity analyses as described in the ODMAP protocol. Results showed no qualitative difference with the results presented in the main text (see Figure   S5.12-S5.16).

| Estimation of index correlation
We computed pairwise Spearman's rho coefficients (ρ) to estimate the spatial congruence among our indices. We tested for the significance of these relationships with corrected degrees of freedom accounting for the positive spatial autocorrelation among observations (Clifford et al., 1989;Haining, 1991 Figure S6.17). It is still widely used in similar large-scale conservation assessments, which allows for standardization and comparative analyses (e.g. Allan et al., 2019;Di Marco et al., 2018;Elsen et al., 2020;Maron et al., 2020;Tucker et al., 2018). The index combines eight variables that measure direct human impacts on the environment, namely (1) human population density, (2) night-time light pollution, (3) extent of built environments, (4) crop land cover, (5) pasture land cover, and (6) proximity to railways, (7) to major roadways and (8) (Brooks et al., 2006). Conversely, we characterized refuge areas as communities within hotspots with the lowest levels of pressures. Refuges should be prioritized for proactive conservation, with the goal of preserving these diverse areas from future anthropogenic threats (Brooks et al., 2006), providing shelter for a portion of ithomiine biodiversity. Finally, we mapped risk areas and refuge areas for four indices selected to represent our classes of highly correlated indices, namely (1) species richness, (2) mean species geographic rarity, (3) mimicry richness and (4) mean mimicry geographic rarity.

| Index maps
The Eastern slopes of the Andes appeared as the primary hotspot of Ithomiini taxonomic, phylogenetic and mimetic diversity, especially between 500 and 2500 m (Figure 3a, c, d, g). We estimated that some quarter-degree grid cells (hereafter referred to as communities) may harbour as many as 120 species, representing up to 28 mimicry rings, especially in Ecuador and Peru. These species totals partly represent alpha-diversity and partly different habitats contained within single quarter-degree grid cells. The Atlantic Forest and the highlands of Central America appeared as secondary hotspots but fall far behind in terms of numbers of species, mimicry richness and phylogenetic diversity. Conversely, we estimated species and mimicry richness,  (Faith, 1992

| Correlation among indices
All indices were significantly positively correlated (Table S10. Table S10.27). Species richness appeared to be a very strong predictor of Faith's phylogenetic diversity (Tables S10.26 and S10.27: ρ = 0.996, t = 84.0, Clifford's df = 56.7, p < 0.001), and mean mimicry ring size (Tables S10.26 and S10.27: Mimicry richness also correlated strongly with species richness (Tables S10.26 and S10.27: ρ = 0.934, t = 19.7, Clifford's df = 56.6, p < 0.001), but it was less strongly correlated with the other indices of the first group (Table S10.26: ρ = 0.854 in average). Moreover, the relationship between species richness and mimicry richness was not strictly linear: some communities with the highest number of mimicry rings are not the most speciose ( Figure S10.24c). Since this pattern can lead to differences in hotspot identification, we attributed mimicry richness to a second class of indices on its own (Figure 4).
Geographic rarity indices (species and mimicry) were more closely correlated with each other (Tables S10.26 and S10.27: ρ = 0.657, t = 7.05, Clifford's df = 65.5, p < 0.001) than with any other indices. However, they were less correlated with each other than the first group of indices. As such, they formed a third and fourth class of indices (Figure 4). They were nonetheless moderately correlated with species richness (Tables S10.26 and S10.27: ρ = 0.473, t = 4.34, Clifford's df = 65.4, p < 0.001 for mean species geographic rarity; ρ = 0.606, t = 5.98, Clifford's df = 61.5, p < 0.001 for mean mimicry ring geographic rarity). Indeed, species-rich communities tended to present high mean geographic rarity values, while species-poor communities exhibited the entire range of relative levels of species endemism (Figure S10.24a and S10.24b). Similarly, communities with high mimicry richness showed high mean mimicry geographic rarity, while communities with few mimicry rings could exhibit the entire range of relative levels of mimicry endemism ( Figure S10.24d).
Correlations including the four additional indices computed (namely species Shannon's diversity, mimicry Shannon's diversity, Evolutionary Distinctiveness and the maximum mimicry ring size) supported the classification in four classes of indices and can be found in Appendices 9 and 10.

| Threat and refuge maps
Our assessment of current anthropogenic threats on Ithomiini diversity hotspots showed that the northern Andean cordilleras combine high taxonomic and mimetic diversity with high levels of human im-  (Figures 5b and 6b; 13% of hotspots compared to the expected 25% overlap).
We estimated mean species geographic rarity hotspots to be relatively more threatened than the other facets of Ithomiini diversity, and also to deviate positively from that expected from a random distribution of anthropogenic threats (Figures 5c-d and   6). This trend remained detectable when exploring other thresholds to define the hotspots (Figure S11 Hotspots of mimicry richness showed similar patterns to species richness due to the relatively high spatial congruence between the two indices (Figures 4 and 5). The main difference was that the Atlantic Forest did not rank as a hotspot for mimicry diversity and therefore reduced perceived threat levels on mimicry richness hotspots compared to species richness hotspots (Figures 5a, e and 6a; 15.8% for mimicry richness and 23.9% for species richness vs. 25% expected overlap).
Likewise, we estimated hotspots for mean mimicry ring geographic rarity to face high levels of threats in the Andes, Central America and the coastal part of the Atlantic Forest. They also extend moderately to the relatively less threatened part of the Upper Amazon, along the course of the Amazon River (Figure 5gh). However, the top 5% hotspots found in the Andes and on the Pacific coast coincided with very few potential refuge areas (Figures 5h and 6b; 1.5% and 0% vs. 25% and 5% expected overlaps, respectively, for the low and very low levels of threats) while being more threatened than expected from a random distribution of anthropogenic threats (Figure 6b; 34.4% and 6.7% vs. 25% and 5% expected overlaps).

| Spatial congruence of the facets of Ithomiini diversity
In our integrative approach to mapping Ithomiini diversity, we found that estimated species, mimicry and phylogenetic diversity indices are strongly correlated across the Neotropics. All indices peaked on the eastern slopes of the Peruvian and Ecuadorian Andes, and in the Upper Amazon region, while the Atlantic Forest and Central American mountains appeared as secondary richness hotspots (Figure 3a, c, d). We also uncovered relatively low levels of biodiversity in the Llanos, the Guyana Shield and the Cerrado savannas. As is common in stack-SDM procedures, these predic-  (Chazot et al., 2018(Chazot et al., , 2019Chazot, Willmott, Condamine, et al., 2016;De-Silva et al., 2016Elias et al., 2009). Ithomiini are strongly specialized on their larval hostplants, and hostplants are likely a limiting resource (Drummond III & Brown Jr, 1987;Willmott & Mallet, 2004). The diversity of Solanaceae, on which most Ithomiini feed as larvae, also peaks in the Andes and the Upper Amazon (Knapp, 2002;Ulloa Ulloa et al., 2017), thereby potentially enabling greater local Ithomiini diversity. This apparent spatial correlation between species diversity in ithomiines and their hostplants is consistent with a hostplantmediated adaptive radiation scenario (Willmott & Freitas, 2006). Finally, the topological complexity of the Andes and adjacent foothills creates high variability in abiotic conditions, habitat and vegetation types (Osborne, 2012), which generates species turn-over, mimicry shifts and fosters vicariant speciation (Chazot et al., 2014;Elias et al., 2009;Jiggins et al., 2006). Therefore, the tropical Andes represent the primary hotspot of alpha-diversity (Figure 3a), but also beta-diversity with high local endemicity (Figure 3b and h) Brown Jr., 2005;Brown Jr. & Freitas, 2002;Sombroek, 2000).
Conservation efforts focused on hotspots of taxonomic diversity alone may not necessarily be effective for conserving a large fraction of species, or other aspects of biodiversity (Devictor et al., 2010;Godoy-Bürki et al., 2014;Williams et al., 1996;Williams & Humphries, 1994;Zupan et al., 2014). As such, areas that host rare and endangered biological features should also be considered (Prendergast et al., 1993;Lawler et al., 2003;e.g

. Alliance for Zero
Extinction: Parr et al., 2009; Key Biodiversity Areas: KBA Standards & Appeals Committee, 2020). Our mean geographic rarity indices provide a useful tool to predict areas of high endemicity that may differ from diversity hotspots. Still, geographic rarity indices appeared positively correlated with species and mimicry richness at the continental scale ( Figure S10.24a and S10.24d). Species-rich communities that host mostly species with small ranges were found in the Andean and Central American mountains (Figure 3b). These regions harbour steep environmental gradients (Osborne, 2012) enabling strong hostplant turn-over (Knapp, 2002), which may limit ithomiine species ranges. Yet, some species-poor communities also coincide with high levels of species and mimicry rarity. Such communities are found in the outer edges of the global distribution of Ithomiini, especially on the Pacific slopes of the Andes (Figure 3b, e, h). There, a strong environmental gradient and geographic barriers such as the Pacific Ocean, the Peruvian coastal desert and the Central Andes could explain the restricted ranges of the few resident species and mimicry rings. Moreover, unusual environmental conditions found at the outer edges of the Ithomiini range may select for specific lineages and mimicry patterns due to environmental filtering.
Overall, Ithomiini biodiversity hotspots appear fairly congruent with biodiversity hotspots based on vascular plant species richness and endemism (Myers et al., 2000). As such, the tropical Andes stand out as the primary hotspot for the multiple facets of Ithomiini biodiversity. Secondary hotspots of Ithomiini diversity, namely the Atlantic Forest, Central American mountains and the Pacific slopes of the Andes, also coincide with areas previously recognized as biodiversity hotspots (Myers et al., 2000). Moreover, our inferred pat-  I G U R E 5 Maps of risk and refuge areas for four predicted Ithomiini diversity hotspots. Only hotspots are displayed, defined as the top 25% (Panels a, c, e, g, on the left) or top 5% (Panels b, d, f, h, on the right) of highest-ranking communities for each index. Anthropogenic threat levels are based on the Human Footprint index, classified within quantiles 5% and 25% for lower and upper end of the distribution. Risk areas (in red) represent areas of overlap between high threat zones and hotspots. Refuge areas (in blue) represent areas of overlap between low threat zones and hotspots. Hotspots not falling into areas of high or low threat levels are displayed in light colour.   et al., 2003). Among insect taxa, the overall biodiversity patterns of Ithomiini are consistent with those of Cicindelinae beetles (Pearson & Carroll, 2001), Adelpha butterflies (Mullen et al., 2011), Nymphidiina butterflies (Hall, 2018) and Heliconiini butterflies (Rosser et al., 2012). Conversely, the distribution of bees in the Neotropics presents an opposite trend, with higher richness per area reached in the Chaco regions, Caatinga, dry Southern Andes and Atlantic forest, reflecting the great success of this group in xeric and seasonal habitats (Orr et al., 2021). Overall, these results support the role of ithomiine butterflies as suitable flagship indicator group for Neotropical butterfly diversity and reinforce the position of the tropical Andes as the flagship region for biodiversity conservation in general and insect and butterfly conservation in particular.  Trullas et al., 2007;Dufour et al., 2018). As such, further research is still needed to better understand the selective advantages of these transparent wings that shape the biogeography of mimicry patterns in ithomiines.

| Distribution patterns of mimetic interactions
Altogether, mimicry richness is expected to follow species richness since more species provide greater opportunities to harbour different wing patterns. In parallel, mimicry fuels species richness by limiting the exclusion effect of competition among co-occurring co-mimetic species (Gross, 2008). However, aposematic signals are predicted to converge locally due to positive frequency-dependent selection incurred by predators (Müller, 1879). Therefore, mimicry richness should increase more slowly than species richness and plateau when all ecological niches are occupied and a (set of) wing pattern(s) already dominates each niche (Joron & Mallet, 1998). The relationship between Ithomiini mimicry and species richness was positive ( Figure 4), but it appeared only slightly saturated (S10.24c).
This suggests that even in species-rich communities, there is some free ecological space, or that the effect of selection for wing pattern convergence is weaker than thought in Müllerian mimetic communities. For instance, the high numbers of mimicry rings found in the most speciose communities in the Ecuadorian and Peruvian tropical Andes may arise because the steep environmental and altitudinal gradients in these regions create a small scale mosaic of zones within each grid cell, hosting locally adapted species and mimicry rings with fuzzy limits (Sherratt, 2006;e.g. altitudinal bands;Chazot et al., 2014). This dense spatial structuring facilitates the recurrent permeation of species and mimicry rings from adjacent zones that may not represent local adaptive peaks of the available niches. Yet, such species could persist, rescued by recurrent immigration (Brown Jr. & Freitas, 2002;Joron & Iwasa, 2005), thereby fuelling local mimicry richness. From a conservation point of view, preserving high species richness should also ensure the preservation of mimicry richness.
We estimated that rich communities may not only harbour the highest number of species but also the highest proportion of mutualistic interactions in the ithomiine butterfly communities, with the largest mimicry rings on average (Figures 3a and f, 4). Thus, species in those communities might be better protected against secondary extinctions that would result from the loss of mutualistic interactions associated with the extirpation of their co-mimetic species.

| Threats and refuges for Ithomiini diversity hotspots
In this study, we highlighted areas of high risks for biodiversity loss, of priority interest for reactive conservation to reduce high anthropogenic pressures on biodiversity. In parallel, we defined refuge areas, of priority interest for proactive conservation to provide shelters for biodiversity from human pressures (Brooks et al., 2006). As such, hotspots for species and mimicry richness, and phylogenetic diversity, located in the Andes and the Upper Amazon, face contrasting situations. While the Upper Amazon has some of the most intact ecosystems in remote areas, the Andes, particularly the rich communities in the western foothills in Ecuador and the three Andean cordilleras in Colombia, are facing high levels of human impacts (Figures 5a, b, e, f and 6). The second diversity hotspot, the Atlantic Forest, is also of great concern, demonstrating the highest level of human pressures and relentless fragmentation of its forested habitats (Ribeiro et al., 2009). Communities with geographically rare species and mimicry rings are found mostly in Andean and Central American mountain ranges, coinciding with areas of high human impacts (Figure 5c, d). Their situation is of particular concern since species with small distribution ranges are known to face higher risks of extinction Cardillo et al., 2008;Purvis et al., 2000), thereby also impacting the narrowly distributed mimicry rings they represent.  (Ohlemüller et al., 2008). Yet, many tropical species, with typically narrow niches and slow niche evolutionary rate, are suspected to lag behind the shift of their climatic envelope (Jezkova & Wiens, 2016). In the case of Ithomiini, which rely on local mutualistic interactions with co-mimics and host plants, the threat of community disassembly due to climate change is even more profound (Sheldon et al., 2011;Toby Kiers et al., 2010). Mountain habitats are particularly under threat from human activities, with high deforestation rates due mostly to the competition for arable lands (Armenteras et al., 2017). Even where human population density is low in remote mountain regions, natural habitats may come under threat from road-building and mining operations (Bax et al., 2019;Sonter et al., 2017).
Remote portions of the Upper Amazon forest may currently act as refuges for a fraction of Ithomiini diversity. Yet, even within protected areas, landscape-level changes can impact insect faunas (Hallmann et al., 2017Salcido et al., 2020). The Upper Amazon remains largely exposed to climate change, notably increases in temperature and drought intensity (Malhi et al., 2008;Nobre et al., 2016), and to deforestation threats (Carvalho et al., 2019;Escobar, 2020).
Thus, the potential refuge areas we have mapped represent only the currently less threatened areas of Ithomiini diversity hotspots, but do not guarantee the conservation of all the biodiversity facets they currently host, especially in the face of global changes. The next avenue for research is therefore to model the effects of climate change and future land-use changes on the patterns of Ithomiini diversity to refine conservation perspectives in a changing world.

ACK N OWLED G EM ENTS
The authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest, or nonfinancial interest in the subject matter or materials discussed in this manuscript. MD is financed by the French Ministry of Research (MENSR). We thank the museum curators who allowed us to

CO N FLI C T O F I NTE R E S T
The authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest, or nonfinancial interest in the subject matter or materials discussed in this manuscript.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13455.

DATA AVA I L A B I L I T Y S TAT E M E N T
All R scripts are available on GitHub at https://github.com/MaelD ore/ithom iini_diver sity. Species-grid-cell records and mimicry clas-

B I OS K E TCH
The research team has a long-standing interest in ithomiine 'clearwing' butterflies, spanning systematics, diversification patterns and processes, the role of host plants and mimetic interactions from micro-to macroevolutionary scale, community ecology and phylogenetics, and historical biogeography. This paper is the result of over 30 years of collaborative efforts to collect, describe and classify Ithomiini butterflies and their mimicry patterns in order to unravel the mechanisms at play in species distribution, community assemblages, and mimicry patterns and niche evolution. The first author, Maël Doré, is a PhD student interested in the interplay between community ecology, phylogenetics and macroecological patterns in tropical environments.
To this end, he employs various tools ranging from systematics to species distribution models and comparative phylogenetic methods.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found in the online version of the article at the publisher's website.